Risk Modeling to Reduce Monitoring of an Autoantibody-Positive Population to Prevent DKA at Type 1 Diabetes Diagnosis

Abstract

Context

The presence of islet autoimmunity identifies individuals likely to progress to clinical type 1 diabetes (T1D). In clinical research studies, autoantibody screening followed by regular metabolic monitoring every 6 months reduces incidence of diabetic ketoacidosis (DKA) at diagnosis.

Objective

We hypothesized that DKA reduction can be achieved on a population basis with a reduced frequency of metabolic monitoring visits. We reasoned that prolonged time between the development of T1D and the time of clinical diagnosis (“undiagnosed time”) would more commonly result in DKA and thus that limiting undiagnosed time would decrease DKA.

Methods

An analysis was conducted of data from TrialNet's Pathway to Prevention (PTP), a cross-sectional longitudinal study that identifies and follows at-risk relatives of people with T1D. PTP is a population-based study enrolling across multiple countries. A total of 6193 autoantibody (AAB)-positive individuals participated in PTP from March 2004 to April 2019. We developed models of progression to clinical diagnosis for pediatric and adult populations with single or multiple AAB, and summarized results using estimated hazard rate. An optimal monitoring visit schedule was determined for each model to achieve a minimum average level of undiagnosed time for each population.

Results

Halving the number of monitoring visits usually conducted in research studies is likely to substantially lower the population incidence of DKA at diagnosis of T1D.

Conclusion

Our study has clinical implications for the metabolic monitoring of at-risk individuals. Fewer monitoring visits would reduce the clinical burden, suggesting a path toward transitioning monitoring beyond the research setting.

Since the early 1990s, research studies from multiple countries have demonstrated that the presence of multiple autoantibodies (AAB) is so strongly predictive of clinical disease that this state has been defined as an early stage of type 1 diabetes (T1D) (1). These research studies longitudinally followed AAB-positive (AAB+) individuals at risk for T1D until the time of clinical diagnosis and have generally involved assessments of glucose tolerance every 6 to 12 months to evaluate progression to clinical diabetes (2–7). Such screening and monitoring studies markedly decreased the incidence of diabetic ketoacidosis (DKA) at the time of clinical diagnosis for those in the study. DKA is life-threatening and occurs when individuals have insufficient insulin secretion to use blood glucose; the body instead uses fat for energy, resulting in the production of ketones, which become toxic. The incidence of DKA among those diagnosed with T1D outside research studies varies from 25% to 58%, with higher incidence in the youngest individuals (8–12). This incidence is reduced to as low as 3.3% when AAB+ individuals participate in active monitoring in the context of research studies (2, 8, 13). Metabolic monitoring is not currently standard of care, but such a dramatic effect on DKA is clearly of clinical importance and monitoring programs may also be fiscally important by reducing costs of DKA at diagnosis (14, 15).

The knowledge that DKA can be prevented through close metabolic monitoring, combined with the successful results of the Type 1 Diabetes TrialNet teplizumab prevention study (16), has increased interest in moving T1D risk assessment from research to clinical care. The initial aim of such programs would be to reduce the incidence of DKA and provide a pool of individuals for trials and therapies when available (17). However, 3 steps are key to clinical implementation: 1) validation of AAB measures or screening algorithms in the general population, such as (18); 2) optimization of metabolic monitoring for disease progression among AAB+ individuals to reduce DKA with minimal burden to participants and clinical care teams; and 3) access to specialist care for therapies and trials. Here, we focus on that second, important step: understanding the frequency and timing of monitoring visits needed to reduce DKA using a risk-based monitoring schedule.

The standard monitoring schedule used in most research studies for those at risk of T1D development involves visits every 6 months. Moving from this twice-yearly monitoring in the context of research toward the future goal of monitoring in clinical practice leads to the question of whether a high, twice-yearly monitoring frequency is necessary to impact the incidence of DKA—or whether a reduced number of visits could still be beneficial on a population level. The framework we applied to address this question centered on the concept of undiagnosed time with T1D; specifically, the longer time with unrecognized T1D, the more likely DKA would occur. Using data from TrialNet's Pathway to Prevention (PTP) study, we developed monitoring visit schedules focusing on optimization of undiagnosed time across a population of individuals with islet autoimmunity. We aimed to optimize both the frequency and placement of monitoring visits according to AAB status and age. Since screening for AABs has been most frequently conducted in family members and/or those with genetic risk, which is strongly associated with human leukocyte antigens (HLAs) (5, 19, 20), we also evaluated the effect of high-risk HLA alleles on optimized visit schedules. In short, we developed models to inform decision-makers about optimized monitoring schedules aimed at reducing DKA in the population of individuals with islet AABs.

Materials and Methods

Study Population

Deidentified demographic, metabolic, AAB, and HLA data from TrialNet's PTP study as of July 2019 were used as previously described (6, 21, 22). After informed consent, this study screens relatives of those with T1D for islet autoimmunity as measured by AABs. Those with AABs are followed with regular monitoring visits until development of T1D or enrollment in a prevention clinical trial. Monitoring visits include 2-hour oral glucose tolerance testing (OGTT) as well as glycated hemoglobin A_1c. Diabetes diagnosis was made from 2 consecutive OGTTs with fasting glucose greater than or equal to 126 mg/dL or 2-hour glucose greater than or equal to 200 mg/dL; symptoms of hyperglycemia and random glucose greater than or equal to 200 mg/dL; or unequivocal hyperglycemia in the presence of other signs and symptoms of disease. During the time period from March 2004 to July 2019, the PTP protocol tested for AAB to insulin, GAD65, and IA2 on all participants. Samples that were positive for 1 of these 3 AABs were subsequently retested for islet cell AAB and, since 2012, for AAB to ZnT8.

Statistical Methods

Optimization of monitoring visits leverages known patterns of risk to place monitoring visits at intervals that minimize undiagnosed time for those who progress from AAB positivity to clinical T1D during monitoring. Our intention was to design a monitoring program for children ages 1 to 18 years, and adults ages 18 to 35 years.

For this analysis, we present optimized monitoring within groups defined by AAB status and HLA type. Therefore, we used a fully stratified parametric Weibull survival model to characterize risk within these groups and based optimization on this risk. This model appropriately handles age-dependent effects on risk variables (22), allows for left-truncated, right-censored follow-up, as well as interval-censored events, all features of the data from Pathway to Prevention. Results, such as estimated value of Weibull parameters, hazard rates, and cumulative incidence, are presented with approximate 95% confidence intervals. Analyses were based on complete cases. Further information about the methods used for this analysis and R code can be found in the Supplemental Methods (23). All analyses were performed using R statistical software (version 4.1.1.) (24).

Results

We obtained and analyzed data from the Type 1 Diabetes TrialNet PTP cohort including 6193 individuals, with a mean ± SD age of 17 ± 13 years, who had at least 1 islet AAB present on 2 separate occasions (Table 1). Using these data, we aimed to develop a model and optimize a schedule that could reduce undiagnosed time in populations of pediatric and adult individuals with single and multiple AABs. To do this we modeled risk of diagnosis as a function of age, including other relevant covariates, and used the patterns of risk to identify optimal timing for monitoring visits.

Table 1.

Baseline descriptive statistics overall and by autoantibody status at baseline

	Overall (N = 6193)	Single AAB+ (N = 3790)	Multiple AAB+ (N = 2403)
Type 1 diabetes	868 (14%)	229 (6%)	639 (27%)
Follow-up time, y	1.9 (0.7-4.0)	2.1 (0.8-4.5)	1.5 (0.6-3.2)
Age at baseline, y	17 ± 13	20 ± 14	13 ± 10
Male	2903 (47%)	1649 (43%)	1254 (52%)
Ethnicity
ȃWhite	5268 (92%)	3173 (92%)	2095 (93%)
ȃBlack/African American	181 (3%)	108 (3%)	73 (3%)
ȃMultiracial	118 (2%)	75 (2%)	43 (2%)
ȃAsian	100 (2%)	78 (2%)	22 (1%)
ȃAmerican Indian/Alaska Native	22 (0%)	15 (0%)	7 (0%)
ȃNative Hawaiian/Pacific Islander	10 (0%)	6 (0%)	4 (0%)
Relationship to proband
ȃFirst	5514 (90%)	3339 (89%)	2175 (91%)
ȃSecond	470 (8%)	311 (8%)	159 (7%)
ȃThird	155 (3%)	107 (3%)	48 (2%)
BMI	21 ± 6	22 ± 7	20 ± 6

Data are summarized by N (%), median (25th percentile-75th percentile), or mean ± SD.

Abbreviations: AAB+, autoantibody-positive; BMI, body mass index.

The standard monitoring schedule used in most research studies for those at risk of T1D development involves visits every 6 months. Considering a pediatric population monitored from age 1 to 18 years (inclusive), this standard would involve 34 visits. Given 34 equally spaced visits and the number of individuals diagnosed in the interval, those with a single AAB would experience a uniform undiagnosed time of 0.25 years across each visit interval (Fig. 1A). The average undiagnosed time for the population is thus also 0.25 years. Using the same number of visits, we next calculated an optimized visit schedule for this single AAB+ population (Fig. 1B). Younger children with a single AAB will have more frequent visits and thus a shorter undiagnosed time than older children in the optimized schedule (range undiagnosed time 0.19-0.30 years). However, as the event rate for the single AAB population is uniformly low, the average undiagnosed time for the population does not differ from that for regularly scheduled visits (0.25 years). For the multiple AAB+ pediatric population, 34 evenly spaced visits result in an average population undiagnosed time of 0.26 years (Fig. 1C). Using an optimized schedule with variable timing between visits reduces the average multiple AAB+ population undiagnosed time to 0.20 years (Fig. 1D). In this case, the range of undiagnosed time within intervals varies from 0.09 to 0.54 years (Fig. 1D). Since more cases occur earlier in life, increasing the visit frequency in younger relative to older multiple AAB+ children benefits the general population.

Figure 1.

Expected undiagnosed time using evenly spaced or optimized visit schedules. Expected undiagnosed time (years) for a group of A and B, 1000 single autoantibody-positive (AAB+), and C and D, 1000 multiple AAB+ individuals in a monitoring program covering the pediatric years (ages 1-18 years). Average undiagnosed time for the single AAB+ population (dotted line) when visits are placed every 6 months (even visit spacing, A) or when optimized visit spacing (B). Average undiagnosed time for the multiple AAB+ population (even visit spacing, C; optimized visit spacing, D). Sizes of circles reflect number of cases within the interval and heights reflect average undiagnosed time within interval. These quantities reflect theoretical results treating population parameter values as fixed and known, based on our modeling results.

We next varied the total number of monitoring visits while also applying an optimized approach to determine visit timing. This analysis demonstrates how changing the number of visits affects the undiagnosed time. For the multiple AAB+ population illustrated in Fig. 2, at the extreme, conducting only 5 visits during the period from ages 1 to 18 years would result in a population average undiagnosed time of 17.9 months. Conducting 34 visits yields an undiagnosed time of 2.4 months (identical to the 0.20 years in Fig. 1D) and selecting a visit count of 14 results in less than 6 months of undiagnosed time for this population, representing a reduced visit count of more than half.

Figure 2.

Relationship between number of visits and undiagnosed time given an optimized schedule in a multiple autoantibody-positive (AAB+) pediatric population. Expected undiagnosed time (years; log scale) and number of visits for pediatric multiple AAB+ population. To achieve a population average undiagnosed time of 6 months, 14 visits would be required. Fewer visits (5 visits illustrated as an example) results in almost 18 months of undiagnosed time, while more visits (34 visits illustrated for example) results in only 2.4 months of undiagnosed time.

We stratified into pediatric and adult populations and the presence of single or multiple AABs, since T1D occurs both in children and adults (25, 26) and practice guidelines for routine clinical care differ for pediatric and adult populations (27). Further, while data support the concept that nearly all individuals with multiple AABs will eventually progress to clinical T1D, the rate of progression is strongly age related (28–30). As illustrated in Fig. 3A, in a single AAB+ population monitored from age 1 to 18 years, the estimated cumulative incidence of T1D is stable with a hazard rate of about 0.01 events/person-year. However, over the same period, the hazard rate in the multiple AAB+ pediatric population exhibits much higher event rates in the first 5 years of life. Among the adult population (Fig. 3B), those with a single AAB also have a low and constant hazard rate of approximately 0.01 events/person-year during the period from age 18 to 35 years. The hazard rate in multiple AAB+ adults during this time is stable around 0.07 events/person-year, similar to the pediatric population with multiple AABs as it approaches age 18.

Figure 3.

Hazard rate for type 1 diabetes (T1D) diagnosis within single and multiple autoantibody-positive (AAB+) pediatric (ages 1-18 years) and adult (ages 18-35 years) populations monitored over 17 years. Incidence lines and 95% confidence intervals for A, pediatric, and B, adult cohorts. The hazard rate for single AAB+ population is constant over period of monitoring whereas the hazard rate for those with multiple AABs is greatest at the start of the monitoring period, especially within the pediatric population. The hazard rate in older teens is similar to younger adults (different y-axis scales).

Given the differences in incidence within these populations, we used our model to determine optimal visit schedules that result in an average undiagnosed time of 0.5 years (6 months) separately for the pediatric and adult populations. For the pediatric population with a single AAB, this is achieved with 17 approximately evenly spaced visits (Fig. 4A). In contrast, to achieve the same average undiagnosed time in the pediatric population with multiple AABs the visit schedule is skewed (Fig. 4B). The higher frequency of visits in the younger ages corresponds to the ages of higher disease incidence (illustrated by the blue bars). Since these early visits capture many of the diagnoses, only 14 visits are needed to achieve the same average undiagnosed time for the multiple AAB+ pediatric population. In the adult population, the optimized visit schedule has 18 visits for those with a single AAB (Fig. 4C) and 17 for those with multiple AABs (Fig. 4D). As in the pediatric population, the younger multiple AAB adults have a higher disease incidence and thus frequency of visits than the older adults.

Figure 4.

Cumulative incidence of type 1 diabetes (T1D) with monitoring schedule optimized to achieve approximately an average of 6 months undiagnosed time within each cohort. Cumulative incidence of T1D (with 95% confidence intervals) with optimized visit schedule for monitoring between ages 1 and 18 years for single and multiple autoantibody-positive (AABs) (pediatric cohort A, B) and for monitoring between ages 18 and 35 (adult cohort C, D). The optimized visit schedule is determined by setting the expected undiagnosed time as not greater than 6 months. Tick marks along the x-axis have been placed at these chosen visits. The vertical shaded bars along the x-axis represent the incidence within the visit window. A nominal 6 months of average undiagnosed time is achieved with A, 17 visits for the pediatric single AAB+ population; B, 14 visits for the pediatric multiple AAB+ population; C, 18 visits for the adult single AAB+ population; and D, 17 visits for the adult multiple AAB+ population.

The aforementioned optimized visit schedules to achieve an expected undiagnosed time of less than or equal to 6 months would require a clinician to use only age and AAB status to stratify a patient to a monitoring schedule. However, some populations tested for AABs may also have HLA results available (5, 20). We thus evaluated whether HLA status would affect either the number or timing of visits in an optimized monitoring schedule, aiming for the same undiagnosed time within a cohort. We first evaluated the hazard rate for individuals with either of the high-risk alleles (HLA DR3 or DR4). Fig. 5 illustrates that the presence or absence of these HLA alleles affects risk primarily among the youngest individuals with a single AAB as the CIs overlap for the multiple AAB+ population hazard rates according to HLA status. We then compared the schedule between those with and without HLA DR3 or DR4 in pediatric and adult cohorts with single and multiple AABs. HLA status had little effect on the optimized visit schedule to achieve a population average undiagnosed time of 6 months in either the pediatric or adult populations with single or multiple AABs (Fig. 6).

Figure 5.

Type 1 diabetes (T1D) risk for single and multiple autoantibody-positive (AAB+) individuals with and without human leukocyte antigen (HLA) DR3 and/or DR4. Hazard rate and 95% confidence intervals for development of T1D among single and multiple AAB+ individuals ages 1 to 50, with and without HLA DR3 and/or DR4.

Figure 6.

Cumulative incidence of type 1 diabetes (T1D) with monitoring schedule optimized to achieve approximately an average of 6 months undiagnosed (UnDx) time within multiple autoantibody-positive (AAB+) pediatric and adult populations with and without HLA DR3 and/or DR4. Cumulative incidence of T1D (with 95% confidence intervals) with optimized monitoring schedule within each cohort. Multiple AAB+ pediatric cohort A, without, or B, with HLA DR3 and/or DR4. Multiple AAB+ adult cohort C, without or D, with HLA DR3 and/or DR4. The optimized visit schedule is determined by setting the expected undiagnosed time as not greater than 6 months. Tick marks along the x-axis have been placed at these chosen visits. The total number of visits and actual average undiagnosed time for each cohort is noted on each panel. The vertical shaded bars along the x-axis represent the incidence within the visit window.

Discussion

Decades of research studies in multiple countries have firmly established that individuals with multiple AABs are expected to progress to clinical T1D. Furthermore, these studies consistently demonstrate that AAB screening in conjunction with regular glucose tolerance monitoring can markedly reduce the incidence of DKA at clinical diagnosis, as most individuals are identified via glucose tolerance testing when they have few if any symptoms of hyperglycemia (8). Monitoring visits in these studies almost always occur every 6 months. Translating this reduction in DKA to a clinical practice setting raises the question as to whether a different monitoring schedule can also achieve DKA reduction in an AAB+ population, with concomitant reduction in burden on AAB+ individuals and their providers.

To address this question, we considered a framework whereby the goal was to minimize the average amount of undiagnosed time for the AAB+ population, reasoning that prolonged time between the development of diabetes and when it is clinician-recognized would associate with increased incidence of DKA. This reasoning is supported by a recent report from The Environmental Determinants of Diabetes in the Young (TEDDY) study illustrating that DKA primarily occurred in those who did not attend routine monitoring visits on schedule (31). Using data from TrialNet's PTP study, we modeled diagnosis of disease across ages 1 to 50, and from this model characterized pediatric and adult populations with single or multiple AABs over an 18-year period for each group. Recognizing a clear trade-off between the number of visits and the amount of undiagnosed time, we determined an optimized visit schedule for each cohort that would limit the average undiagnosed time to not more than 6 months, though the algorithm can also be used to target other levels of average undiagnosed time. As an extreme example, selecting an average of 18 months of undiagnosed time in the pediatric multiple AAB+ population would require only 5 visits during a monitoring period from age 1 to 18 years. Here, we targeted a 6-month average undiagnosed time to mimic the schedule of twice-yearly visits commonly used in research studies that have demonstrated dramatic reductions in DKA at diagnosis. As illustrated, this level of undiagnosed time can be achieved with only 17 or 18 visits across 17 years in the single AAB+ pediatric or adult cohorts, respectively, and with 14 or 17 visits in the same age populations but with multiple AABs. This analysis implies that conducting approximately half the number of visits (14-18 visits for our schedule vs 34 for research studies) and using modified visit placement over a 17-year monitoring period is likely to be effective in substantially reducing the incidence of DKA to the levels seen in research studies both for children and adults. Adding HLA information has only a small effect on the placement and number of visits, for single or multiple AAB+ pediatric or adult cohorts. Of note, we evaluated only the effect of HLA DR3 or DR4; if other HLA risk groups were modeled, other variations in visit schedules may occur. This result is consistent with multiple reports that HLA primarily influences the development of autoimmunity and not progression from autoimmunity to diabetes (32–35).

While the model provides some confidence that an optimized schedule would reduce DKA if applied within the identified population, testing of the model in other data sets would provide additional support, particularly since the TrialNet PTP study includes only relatives of those with T1D. However, since the rate of progression of multiple antibody positivity in other studies that identify individuals either through screening those with selected HLA risk alleles (33) or via true population screening at young ages (2) does not appear to be different from those who were screened because they were relatives, we expect that an optimized schedule of those who are not relatives would be similar. If applied in “real-world” clinical practice, the incidence of DKA should be carefully tracked to confirm the utility of the monitoring schedules, including the possibility that even less frequent testing would be similarly effective. It is important to note that this analysis used TrialNet data, in which study visits include symptom awareness and education as well as metabolic testing. It is not clear which of these factors influences the reduction of DKA seen in research studies; here we address only the monitoring visit schedule and not the components of monitoring.

This work aimed to optimize monitoring visit schedules to reduce the incidence of DKA in the overall population from that seen under the existing standard of care, which currently does not incorporate either AAB screening or monitoring. As in follow-up monitoring in current research studies, such an approach will not eliminate all DKA, rather the model should improve overall outcomes for the population. Our approach places monitoring visits where progression events are most likely to happen in a population. This occurs both when risk is high and when a considerable proportion of the population is at risk. As a result, a schedule that limits the population average undiagnosed time will shift monitoring visits away from ages at which a lower number of patients progress. In a high-risk population this could mean that, for the small fraction remaining diagnosis free at older ages, monitoring could be less frequent.

In contrast to the population approach evaluated here, another approach under development is more personalized, whereby an AAB+ individual's parameters are used to determine their individual time-dependent risk of progression to T1D. This is akin to work conducted in peanut allergy, where prognostic biomarkers are used to model probabilities of clinical outcomes (eg, severe reactions to peanut), which a clinician can then use to personalize risk mitigation and patient care strategies (36). This approach is sensitive to an individual's risk at a given time and therefore may be more appealing to an individual and their clinicians but may create challenges for health care systems induced by a more complex treatment algorithm.

Screening for AABs followed by monitoring for disease progression in the context of research studies can reduce incidence of DKA at the time of T1D diagnosis. Yet development of clinical practice guidelines must also consider financial and operational components of implementing a screening and monitoring program within health care systems, minimizing the burdens while maintaining the potential benefits (37–39). Having introduced the concept of undiagnosed time as a key variable contributing to DKA, our work suggests optimal visit schedules for monitoring of the defined AAB+ populations that, even if slightly simplified for feasibility (e.g., 4 years rather than 4.1 years), should reduce the incidence of DKA at diagnosis if applied in clinical practice. Since the highest incidence of disease occurs in younger individuals, the frequency of visits of younger patients is increased and that of older individuals is consequently reduced to achieve a targeted population-based average undiagnosed time. This population-based approach can inform insurers and policy makers aiming to develop and implement clinical practice guidelines both to identify and monitor individuals at risk for T1D. Reducing the number of monitoring visits while still achieving clinically important reductions in DKA incidence decreases the burden on individuals and health care systems while sustaining the potential benefits, thus enhancing the rationale to implement screening and monitoring for T1D risk in clinical practice.

Conclusions

The risk of T1D is heterogeneous across a population and varies by age, AAB positivity, and other factors. Using patterns of risk within an AAB+ population to minimize average time spent with undiagnosed disease can inform a monitoring strategy for developing clinical practice guidelines aiming to substantially reduce the incidence of DKA at T1D diagnosis.

Abbreviations

AAB

autoantibodies

DKA

diabetic ketoacidosis

HLA

human leukocyte antigen

OGTT

oral glucose tolerance testing

PTP

TrialNet's Pathway to Prevention

T1D

type 1 diabetes

TEDDY

the environmental determinants of diabetes in the young

Financial Support

This work was supported by the Helmsley Charitable Trust (research grant Nos. 2103-05008 and 2210-05590 to C.J.G.). The Type 1 Diabetes TrialNet Study Group is a clinical trials network funded by the National Institutes of Health (NIH) through the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), the National Institute of Allergy and Infectious Diseases, and The Eunice Kennedy Shriver National Institute of Child Health and Human Development, through the cooperative agreements U01 DK061010, U01 DK061034, U01 DK061042, U01 DK061058, U01 DK085461, U01 DK085465, U01 DK085466, U01 DK085476, U01 DK085499, U01 DK085509, U01 DK103180, U01 DK103153, U01 DK103266, U01 DK103282, U01 DK106984, U01 DK106994, U01 DK107013, U01 DK107014, UC4 DK106993, UC4DK117009. Neither NIDDK nor the Helmsley Charitable Trust were involved in the design, analysis, or interpretation of data for this study; writing the report; and neither imposed any restrictions regarding the publication of the report.

Author Contributions

C.O. obtained data, performed analysis, and wrote sections of manuscript; A.Y. assisted with obtaining data; A.Y. and H.T.B. reviewed and assisted with analysis; all authors contributed to study design and interpretation of the data, as well as reviewing and editing the manuscript. C.J.G. obtained funding for the project, and is the guarantor for this study and as such accepts full responsibility for the work, had access to the data, and controlled the decision to publish.

Disclosures

The authors have no conflicts of interest to disclose. Outside the submitted work, C.J.G. has served on advisory boards for Merck, Viela Bio, Enthera Pharma, GentiBio, and Altheia Science, and has received research support from Bristol-Myers Squibb and Janssen, and gave an educational presentation to AstraZeneca. C.S. serves on an advisory board for Vertex Pharmaceuticals.

Data Availability

The data used for this study come from Type 1 Diabetes TrialNet's PTP study and are available at the NIDDK Central Repository. The majority of the analytical code for this study is presented in the Supplementary Materials (23); additional code available on reasonable request.